# Usage and the license

An enterprise agreement commits a volume rather than buying a monthly allowance, so the administrator's question is not "have I run out" but "where are we against what we signed". This page is where that number comes from and what is inside it.

## The unit

One dispatched action. An assistant asks atmon to do a thing in a connected account, atmon does it, and that is one.

The count is taken from the receipts, which means the number an administrator reads and the number on an invoice come from the same rows rather than from two counters that can disagree.

## What is never counted

Four things, and all four are deliberate.

**A call your rules refused.** Governance that costs money to enforce is governance somebody eventually turns off.

**A call parked waiting for an approver.** It counts when it runs, if it runs, and never twice.

**Searching for an action and reading its description.** Finding the right thing is not doing it, and charging for the looking would push an assistant toward guessing.

**Model tokens.** When a job uses a model endpoint you registered, you pay whoever provides the model. atmon does not resell it.

Viewer seats are free on every tier and unlimited, so read-only access for a risk or finance team costs nothing.

## Reading usage

In the console, on the governance surface, which is where an administrator's sign-in already lands.

Over the API, `GetUsage` aggregates a window of receipts grouped by day, by app, or by action:

```json
{ "from": "2026-07-01T00:00:00Z", "to": "2026-08-01T00:00:00Z", "groupBy": "day" }
```

Each row carries the calls in that bucket and the split of them: succeeded, failed, and denied. Those three partition the total, so a month where the number moved is a month you can immediately say why about, rather than one that only shows a number that moved.

Usage is scoped to the project the calling key belongs to, and a request cannot ask about another project. Reading across an organization is an administrator's read in the console, not a wider key.

## Against the license

The commercial shape of an enterprise agreement, the committed volume, the seats and the term are on the [pricing page](https://enterprise.atmon.ai/pricing) and in your contract. What the software does with them is this.

An enterprise deployment does not have a monthly allowance that runs out. What it has is a committed volume on a contract and a live count of what has actually been dispatched, so the reconciliation is arithmetic rather than an audit.

Ceilings are yours to set and are separate from the commitment. A budget ceiling parks work rather than denying it: when a project passes the limit you gave it, further calls stop and wait rather than failing, so the thing that stops is spending rather than the assistant's ability to report that it stopped. Set those per project to whatever the internal chargeback expects.

Three capabilities are entitlements of the enterprise agreement rather than settings: the on-prem relay, OTel export, and single sign-on. The first two work today. [Single sign-on is not built](./index.md), and the roadmap is where it stands.

## A deployment inside your network

A node on your machines counts the same way, from the same receipts, and reports the same numbers to you. What changes is that the count is a record you hold and read rather than a meter we read, and that no call detail crosses the boundary for us to read it: the reconciliation is a usage record, not your traffic.

The mechanism that carries that record back at the end of a term is not built yet, and the shape of it is a commercial question your agreement answers rather than a setting in the software. Until it lands, the count is the one you read from your own deployment through the API above.